Machine learning can predict mild cognitive impairment in Parkinson's disease

Area of Science:

• Neurology
• Biomarkers
• Machine Learning

Background:

• Parkinson's disease (PD) cognitive decline involves non-motor symptoms and gait alterations.
• Amyloid-beta (Aβ) PET imaging has not consistently linked Aβ plaque deposition to PD with mild cognitive impairment (PD-MCI).

Purpose of the Study:

• To identify significant features for predicting PD-MCI using a machine learning approach.
• To evaluate the utility of clinical, gait, and Aβ PET data in PD-MCI prediction.

Main Methods:

• Machine learning models were developed using clinical variables, non-motor symptoms, freezing of gait (FOG), and gait analysis data.
• A subgroup of patients also underwent Aβ PET imaging, which was incorporated into a second model.
• Patients were classified as PD-MCI or non-MCI (noPD-MCI) based on neuropsychological testing.

Main Results:

• The study included 75 PD patients (33 PD-MCI, 42 noPD-MCI).
• PD-MCI patients were older and exhibited worse gait patterns, including increased dynamic instability and reduced step length.
• Machine learning Model 1, using clinical and gait features, achieved over 80% accuracy and specificity; Model 2, including PET data, yielded disappointing results.

Conclusions:

• Machine learning models incorporating clinical features and gait variables accurately predict PD-MCI.
• Amyloid PET imaging did not enhance the predictive accuracy for PD-MCI.
• Gait parameters analyzed via data mining may serve as reliable surrogate biomarkers for PD-MCI.